UNIV OF MARYLAND - HCIL TECHNICAL REPORT (AUGUST 2010)

Innovation Trajectories for Information

Visualizations: Comparing Treemaps,
ConeTrees, and Hyperbolic Trees
Ben Shneiderman, Senior Member, IEEE, Cody Dunne, Puneet Sharma, and Ping Wang
Abstract This paper reviews the trajectory of three information visualization innovations: treemaps, conetrees, and hyperbolic
trees. These three ideas were first published in the early 1990s, so we are able to track academic publications, patents, trade
press articles over almost two decades. We describe the early history of each approach, problems with data collection from
differing sources, appropriate metrics, and strategies for visualizing these longitudinal data sets. This paper makes two
contributions: (1) it offers the information visualization community a history of how certain ideas evolved, influenced others, and
were adopted for widespread use and (2) it provides example of how such trajectories of historical trends can be gathered and
visualized. Guidance for innovators and future analysts is offered.
Index Termsinformation visualization, innovation trajectories, graphical user interfaces, treemap, cone tree, hyperbolic tree.

1 INTRODUCTION

HE healthy outpouring of innovations from the information visualization community has raised important
questions about how to measure the efficacy, adoption,
and durability of these innovations. These measures could
help retrospective analyses that seek to compare several
technologies, but the greatest interest is predictive models
that forecast eventual impact of novel technologies. This
goal may be difficult to attain, but simpler descriptive and
explanatory theories can be helpful to guide future entrepreneurs [21], corporate research managers [1], government funding agency staff [20], and historians of science as
they seek to understand the evolution of technology [4][5].
Another goal is to develop prescriptive theories based on
these metrics, which suggest guidelines and policies for
promoting technological innovations.
Initial questions of efficacy can be partially settled by
empirical testing with traditional controlled experiments
with a few dozen subjects for a few hours that compare
an existing visualization against the innovation [32]. Critics of this approach suggest that short training periods
with standard tasks are insufficient to test innovations
that may require more substantial training, a wider variety of tasks, and even changes to familiar problem-solving
strategies [33] and work processes and practices. These
critics advocate extended case studies with users who
work for weeks and months using the innovative visualization on their own tasks. Longer term measures of efficacy account for other changes such as improved interfaces, better training, simpler integration with other tools,
and a community of like-minded users who are capable
of discussing advanced uses of an innovative visualiza-

tion. Evaluations of efficacy often lead to refinements,

clarification of which tasks are more helped, and provide
a basis for promoting an innovation.
Measuring adoption (also called acceptance or diffusion) beyond the originators is also a challenge as innovations may spread in different ways [30]. Simple measurements in the 2-5 years following the initial presentation of an innovation, include the references in later papers, implementations by multiple open source or commercial organizations, and use by a growing set of users.
Some impacts are more noteworthy than others, such as
solving important problems, sales of commercial products, or inspiring further innovations. However, impacts
may take years or decades to emerge and innovations
may need much transformation or integration with other
ideas to have impact. The longer-term (5-15 years) diffusion through organizations, industries, and countries is
likely to be segmented, as in reaching novice or experienced users, professionals or consumers, and old or
young users. The Technology Acceptance Model [12] focused on perceived usefulness and perceived ease of use,
but later versions incorporated other parameters [37].
Measuring durability (also called sustainability) over
decades is important in understanding why innovations
lose their enthusiastic advocates or satisfied users, how
later innovations replace earlier innovations, and how
innovations become so accepted as to become invisible,
unmentioned, and taken-for-granted.
These measures may be difficult to capture, difficult to
compare, imprecise, and unstable over time. For example,
the number of websites mentioning an innovation only
began to be relevant by the later 1990s, so comparisons

with earlier innovations is difficult. Download counts for

Ben Shneiderman, Cody Dunne, and Puneet Sharma are with the Deparsome software is useful, but sometimes a few hundred
ment of Computer Science and Human-Computer Interaction Lab, Univerdownloads by valued users can be a success while in othsity of Maryland, College Park, 20742.
er situations millions of downloads may be needed to
E-mail: {ben, cdunne, puneet}@cs.umd.edu.
Ping Wang is with the College of Information Studies, University of Maryland, College Park, 20742. E-mail:pwang@umd.edu.

xxxx-xxxx/0x/$xx.00 200x IEEE

claim success.
Another class of difficulties relate to names used for
search engines, digital libraries, or databases. Some innovations have distinct names that are easy to track, while
others have generic names that are harder to search on.
An innovation may be given different names by those
who make refinements and credit may not always be given to the originators. Often an innovation becomes included in a larger product so it is difficult to track.
This paper uses counts of academic papers, patents,
and trade press articles as measures to describe the innovation trajectories of three tree visualization methods:
treemaps, cone trees, and hyperbolic trees. A secondary
measure is the number of citations to key academic papers and patents. We use these measures and other
sources to try to explain their relative successes and give
guidance to promoters of information visualization innovations.
A still more difficult goal is to use early measures to
predict which future information visualization innovations will become successful. This latter goal is aligned
with current U.S. National Science Foundation efforts to
develop a Science of Science, which would help program managers allocate funding more effectively to highpayoff research proposals.
This paper begins with a personal historical review of
tree browsing and visualization methods that focuses on
treemaps, cone trees and hyperbolic trees. Then it discusses how to gather data for tracking the trajectory of
these methods from academic publications to commercial
applications, followed by data visualization to show these
trajectories. The conclusion discusses the difficulty of
such reviews, lessons learned, and future work.

2 TREE BROWSING AND VISUALIZATION

Even early uses of computers, such as Doug Engelbarts
famous 1968 demonstration of his Augment system presented strategies for browsing tree-structured information, often described as hierarchies. Some tree browsers
simply used indented textual representations, which became standard in many directory browsers such as the
Microsoft Windows Explorer. Indented textual representations allow rapid scanning down the list, alphabetical
ordering, and comprehensible expand/contract strategies
to support exploration. The disadvantages include the
need for frequent scrolling as the number of nodes grows
and the difficulty in discovering large subtrees that might
be several levels down.
The appeal of graphical user interfaces encouraged
many developers to create node-link diagrams which became widely used during the 1980s. These visualizations
were readily understandable, whether drawn with the
root node at the top, bottom, left, or right side. With small
trees of 10-50 nodes this strategy was effective, but with
larger trees that had large depth (many levels) or high
branching factor (high fan-out), drawing a complete tree
was impossible, even with megapixel displays. While
panning and zooming facilitated exploration, and overviews helped even more, the node-link diagram had its

limitations.

2.1 Early History

By the early 1990s several research groups developed innovative methods of tree browsing that offered fresh overview and browsing strategies. In each case they developed
a space-limiting approach which ensured that the entire
tree would always be seen on standard displays, thereby
avoiding the need to pan and zoom, but requiring some
new user actions. This paper focuses on three of these tree
browsing methods:
Treemaps used nested rectangles to show tree structure,
producing not only a space-limiting but also a screenfilling algorithm. The area of each leaf rectangle was determined by one of its attributes, and interior rectangles
were sized by the sum of the attribute values of its subtrees and colored by any other attribute value. The original recursive algorithm, slice-and-dice, organized subtrees in a meaningful order (alphabetical, chronological,
etc.), but often resulted in long thin rectangles (high aspect ratios) [31][17]. Improved layouts produced more
square-like aspect ratios (i.e. closer to 1) [41][7], but sacrificed the lexicographic ordering. Bruls et al.s squarified
algorithm caught on widely because it was visually appealing (typically placing large squares in the upper left
and small squares in the lower right), eased selection, and
often permitted better labeling. Still further refinements
offered ordering, low aspect ratios, and some new strategies [6]. The University of Maryland HCIL treemap history page offers numerous links1.
Cone trees used a 3-dimensional layout with the root
node at the top and the first level of nodes were hung
down in a circular layout connected by links to the root
node, thereby looking like a cone [29][28]. Lower levels of
nodes were also laid out in cones all the way down to the
leaf nodes at the bottom of the screen. The appeal of 3dimensional tree structures was strong and the animated
rotation of cones to bring occluded nodes to the front
made for eye-catching demonstrations. Perspective effects, lighting models, and shadows added to the impressive visual appeal. When the root node was placed at the
left side and lower levels to the right, the design was
called cam tree since the appearance was similar to an
automobile cam shaft. Tree-structure traversal and node
name searching were the main emphasis, so node color
and size coding was not discussed. Robertson, inspired
by early scientific visualization researchers, applied the
power of high-end graphics workstations to create abstract data animations based on physical processes. Their
published video produced a vigorous response and great
interest.
Hyperbolic tree browser retained the appealing nodelink visual presentation but placed the root node in the
center with first-level nodes arranged around it in a circle
or oval. Further levels were placed in larger concentric
1

www.cs.umd.edu/hcil/treemap-history/

SHNEIDERMAN ET AL.: INNOVATION TRAJECTORIES FOR INFORMATION VISUALIZATIONS:

COMPARING TREEMAPS, CONETREES, AND HYPERBOLIC TREES

circles or ovals [19], thus preserving a 2-dimensional planar approach. To ensure that the entire tree would be visible, outer levels were shrunk according to a hyperbolic
formula. The compelling aspect of the hyperbolic tree
browser was that users could drag any node to the center,
thereby redrawing the tree in a smoothly animated way
that was innovative, playful, and eye-catching. The developers were inspired by Eschers art work and drew on
the popular notion of fisheye layouts that magnified the
central area while diminishing the peripheral areas. They
adopted the then current terminology of focus+context,
suggesting that the main item of interest was entirely
viewable, while the context remained visible, but reduced
in size. Node labels were readable for nodes in the focus
area while truncated labels were used at the periphery.
Using node size and color to represent node attribute values was introduced in later versions.
Treemaps were invented by the first author of this paper (Shneiderman), while working at the University of
Maryland. He was seeking to solve an immediate problem of understanding disk space usage on a machine
shared by 14 users. Inclined to visual solutions, he tried
many variations until he had the inspirational moment
that led to the compact recursive algorithm. He then
worked with graduate student Brian Johnson to implement the algorithm and refine the design. The original
paper dragged through the journal review and production process, so it was published in 1992, months after the
implementation description appeared at a 1991 conference which had a more rapid turnaround. Cone trees
and the hyperbolic tree browser originated at the Xerox
Palo Alto Research Center (PARC), where a well-funded
team of researchers was developing advanced visual interfaces from as early as 1988. While cone tree authors
Robertson, Card, and Mackinlay were distinct from the
hyperbolic tree browser authors, Lamping, Rao, and Pirolli, they collaborated on other projects and co-authored
related papers. The University of Maryland did not file
any patents for treemaps, but Xerox submitted multiple
U.S. patents on aspects of cone trees (5295243, 5689628,
6088032, and others) and hyperbolic trees (5590250,
5619532, 6300957, and others).
All these researchers were enthusiastic about their innovations and had many opportunities to present their
ideas in journals, at conferences, and in professional talks.
They were all respected and productive researchers,
whose work would draw attention from colleagues. As a
university researcher Shneiderman had the advantage of
graduate students who would implement and evaluate
variations on treemaps, while the Xerox PARC researchers were a capable well-funded group of professionals
who were embedded in a community skilled at producing innovations.

2.2 Proving Efficacy

Proving efficacy of new ideas is often difficult, as the innovators struggle to identify relevant tasks, decide on the
degree of training, choose appropriate subjects, and assess
long-term performance. First versions of new ideas typically require many refinements to reach maturity plus integra-

tion with existing technologies. While the two initial treemap papers did not offer evaluations of efficacy, work was
already underway on two evaluations [35]. The first, using
12 participants, showed performance benefits for treemaps
over UNIX commands for specific directory browsing
tasks. The second, using 18 participants, showed performance benefits for treemaps over paper reports containing
financial data. These modest studies were especially helpful in understanding the strengths and weakness of treemaps, which led to many improvements.
Cone tree developers did not conduct traditional evaluations:
Evaluation of this work [on cone trees] was a topic of
frequent conversations. We had a firm belief that we
needed a more holistic approach to evaluation of the
whole user experience rather than the usability techniques which were in use at the time (which tended to
focus on specific design features, and we believed might
be misleading). Lacking that holistic evaluation approach,
we did no formal user studies. So, the only informal evaluation that was done was based on our own experience as
information workers, and the observations we made
when using the new techniques compared to the existing
ways of looking at the same information structures. (Robertson, personal communication, 3/21/2010).
The initial hyperbolic tree browser paper described an
evaluation of four users with a World Wide Web hierarchy of URLs. The counterbalanced within-subjects design produced subjective preferences that favored the
hyperbolic tree browser compared with a conventional
2-d scrolling browser with a horizontal tree layout.
However, there were no significant performance differences.
Subsequent evaluations of tree searching were conducted by many other researchers. A comparison of several space-filling tools for tree visualization found problems with treemaps but performance improved with experience [34]. However, their tasks emphasized traversal
and nesting levels for which treemaps are known to be
weak, while tasks involving node attribute values were
not included.

2.3 Adoption and Durability

Many sources describe the evolution of novel ideas into
mature technologies and possibly commercially successful
innovations [4][5]. Measuring the efficacy, adoption, and
durability of an innovation remains difficult, but clearly
some ideas such as the World Wide Web have enormous
impact and widespread adoption, while other ideas like pie
menus have a narrower impact and limited adoption. For
the innovations discussed in this paper, the remainder of
this section covers critical incidents, influence on other research groups, commercialization by companies, and visibility in popular media. Section 3 covers citation analysis
for academic papers, patents, and trade publication articles.
As Moore [21] points out innovators face a modest
challenge in gaining early visionary adopters and the
much greater challenges in crossing the chasm to reach an
early majority of the more pragmatic business users. His
arguments are especially relevant to disruptive innova-

tions that require changes to existing practices and ways

of thinking. Novel visualizations still face this challenge
and have the additional burden that many people are
resistant to visualizations compared to textual or tabular
numerical presentations. Educational efforts to promote
visual literacy are spreading as more interactive visual
strategies are developed and as the benefits in corporate
utilization grow. Innovators are typically enthusiastic
about the benefit of their work and they successfuly engage with other innovators who are attracted to novelty.
However, reaching the wider circles of early adopters, the
early majority, and the later majority that Moore describes can be difficult.
Proponents of novel information visualizations frequently report that an effective method is to show potential users their own data or at least familiar data sets using the novel method. This is in harmony with Moores
advice to focus on narrow vertical market segments
where measurable advantages can be shown.
Treemaps were developed for disk directory browsing
which had natural and meaningful variables for size (file
size) and color coding (file type or age). A memorable
incident was a presentation by Shneiderman to the University of Washington Dept. of Computer Science on January 13, 1993. Faculty and student attendees were skeptical about the treemap as a directory browser and even
hostile to the idea of replacing UNIX, DOS, or current
graphical user interfaces. After the talk, Shneiderman
went to their lab and inserted an early demonstration disk
into one of their workstations to show them their disk
directory. The gasps from viewers were immediate they
saw that much of their space was wasted by three copies
of a large compiler that had been installed by separate
users. The University of Maryland did not file patents but
did succeed in securing at least 17 licenses of the software, bringing enough income to support several student
Research Assistants.
While several companies began offering a treemap addition to their directory browser adoption was slow. Jarke
van Wijk and his team developed a free program, SequoiaView, that applied his refinements of squarified and
cushion treemaps. SequoiaView produced hundreds of
thousands of downloads, a popular following, and a spinoff company called MagnaView.
A big increase in adoption and an expansion to a very
different application domain came in 1999 when Martin
Wattenberg working for Smartmoney.com developed
Map-of-the-Market2 to show more than 500 stocks in 11
industry sectors, which were divided into subsectors, all
as leaf nodes at the third level. Wattenbergs clustered
layout, developed independently of the squarified layout,
used size to indicate market capitalization and color to
indicate degree of rise or fall since the previous day.
This free web application which had good labels and
helpful tooltips with links to extensive background data
generated a huge following. However, Smartmoney.com
was only moderately successful with charging a monthly
fee for more sophisticated treemaps or in licensing their
2

www.smartmoney.com/marketmap

software. Shneiderman became a consultant to Smartmoney.com in 1999 and developed a working relationship
with Wattenberg that led to further research results [6].
Another free public website that increased impact was
Marcos Weskamps Newsmap3 which used a treemap to
visualize the data on active news stories as reported by
the Google news aggregator. In April 2004, just after
Weskamp released Newsmap, it generated 114 comments
on Slashdot.org4.
By 2007, Wattenberg, working at IBM Research with
others, produced ManyEyes5, a popular collaborative visualization web site with 20 visualizations including
treemaps. This open tool expanded adoption as thousands of users were able to upload their data and view
the results in treemaps or other visualizations. Also in
2007, The New York Times began using their own interactive treemap software. They continue to use treemaps to
help tell stories such as auto industry sales, financial
trends, budget allocation, and inflation factors.
These and other treemap applications as well as several open source implementations increased the impact and
broadened the adoption, but specific numbers of users are
hard to find. The movement to treemap commercialization was advanced by companies such as The Hive Group
(for which Shneiderman is a Technical Advisor), Panopticon, Lab Escape, Macrofocus, Magnaview, ILOG, and
others listed on the Wikipedia page6. The Hive Groups
CEO, Jim Bartoo reports: Hundreds of organizations
have licensed the companys Honeycomb visualization
software including some of the largest banking, oil & gas,
pharmaceutical, and consumer products companies in the
world With the exception of the economic crisis associated with 2009, The Hive Groups sales increased between 40% and 50% each year since 2005 (personal
communication, 3/28/2010). In addition, the Wikipedia
page lists 16 versions of treemap software in development
libraries in these languages: Adobe Flex, Java, JavaScript,
Silverlight, Perl, Python, Ruby, and SVG. Section 3 reports on the citation and patent history. It may be too early to discuss durability, but numbers of companies, websites, and users continues to grow.
Cone trees produced a strong immediate response with
great interest in the academic community. In conjunction
with the first published paper [29] at the 1991 ACM CHI
Conference they also published a video Information Visualization Using 3D Interactive Animation, which was distributed by the SIGGRAPH Video Review as a videotape to
academic researchers, course instructors, and industrial
laboratory developers. A group at the University of Waterloo created their own cone tree implementation with fewer
3D effects and what they believed to be refinements such as
a better node layout [8]. They conducted a small user test
with five users and three tasks with mixed results but some
guidance for refinements. A variation of the cone tree for
MySQL databases was shown in a 2006 YouTube video by
3

newsmap.jp
news.slashdot.org/article.pl?sid=04/03/31/2332203
5 manyeyes.alphaworks.ibm.com
6 en.wikipedia.org/wiki/List_of_Treemapping_Software
4

SHNEIDERMAN ET AL.: INNOVATION TRAJECTORIES FOR INFORMATION VISUALIZATIONS:

COMPARING TREEMAPS, CONETREES, AND HYPERBOLIC TREES

Daniel Bierwirth7. A commercial application of cone trees

was created by Xerox spinoff XSoft under the name Visual
Recall but it did not reach many users. A demonstration of
the S3 graphics chip used cone trees in their 1997 Visidrive
3D product. No Wikipedia article was found for cone trees.
Hyperbolic trees was first published in a two-page description in a 1994 conference [18], but the fuller description appeared in the 1995 ACM CHI Conference, which
generated excitement and widespread interest [19]. The
original video, Visualizing Large Trees Using the
Hyperbolic Browser, published by the CHI Conference
by way of the SIGGRAPH Video Review, shows examples
of organization charts and web browsing8. Hyperbolic
trees (HT) generated variations from university groups
and was spun off from Xerox PARC as a commercial
product from Inxight. There were commercial and public
uses such as web-based hyperbolic trees for popular web
sites like NASA. Inxight CTO Ramana Rao reports that
they licensed toolkits to implement HT in vendors or
solution providers packages (~100 deals) and several end
user products including an enterprise server product
(~500 servers) that allowed for incorporating HT into internal web apps and a end user product for publishing
site maps on websites (~5000 sites) and maybe 500k users
of a Windows navigation freebie called Magnifind. The
revenue over 10 years was on the order of $30M (personal communication, 3/27/2010).
Hyperbolic trees were renamed Startrees by Inxight,
which was acquired by Business Objects in 2007, which in
turn was acquired by SAP, which continues to offer licensing at its website9. Inquiries to SAP about sales history were not responded to. Rao laments the absence of a
champion at SAP and the declining number of public
web-based uses.
The Wikipedia article on hyperbolic trees10 gives seven
external links including examples that remain on the web
such as the Green Tree of Life from Univ. of CaliforniaBerkeley11. An unidentified developer provided an open
source version called Hypergraph on the Sourceforge
website12. However, strong patent protection by Xerox
PARC raises questions about legality of usage of this code
or the Inxight source code, which was also posted at
Sourceforge with clear warnings.
One enthusiastic writer for an information professionals magazine wrote about The Hype Over Hyperbolic
Browsers [2]: For libraries, this approach could revolutionize subject searching but she cautions that Unfortunately, I predict that it is unlikely we will see hyperbolic
browsers or visually orientated relational databases replacing traditional search engines any time soon. Current
7

www.youtube.com/watch?v=1eO1pgTVu-g
www.youtube.com/watch?v=pwpze3RF55o

TABLE 1
CITATION AND DOWNLOAD COUNTS FOR EACH INNOVATION
Google SchACM DL ACM DL ACM DL CiteSeer
lar Cites Cites
Downloads Downloads,Cites
last 6 weekslast 12 mont

Treemap
ACM TOG
1992 [28]
IEEE Vis
1991 [16]
Cone tree
ACM CHI
1991 [29]
CACM
1993 [28]
Hyperbolic
tree
ACM CHI
1995 [19]
UIST
1994 [18]

764

145

61

373

237

754

155

52

321

164

1132

242

48

450

382

576

126

75

557

227

833

154

18

246

263

48

16

53

76

207

Citation and download counts as of March 20, 2010 for the key academic
papers for each innovation. All three designs show strong and continuing
patterns of interest, with cone trees generating the strongest showing.

conventions for searching and information retrieval are so

ingrained that will take an enormous cultural shift among
information professionals to pave the way for graphically
oriented search tools. This comment, painful to all information visualization researchers, may have an important message about the difficulty in gaining acceptance to
many of our innovations.
Competition over browsing strategies for hierarchies
produced a lively event at the ACM CHI97 conference in
which six teams competed to browse a large hierarchy
[24]. Results were complex but Ramana Rao using the
hyperbolic tree browser clearly stood out. However, the
tasks chosen involved only navigation over named nodes
with no attribute values so the benefits of treemaps size
and color coding were not applicable.
Other evaluations of the hyperbolic tree browser
helped reveal its strengths and weaknesses for specific
tasks [27][26][42]. A survey article covering overview+detail, zooming, and focus+context interfaces discussed tree browsing plus other applications, reviewed
the empirical studies, and provided a principled analysis
based on perceptual psychological issues [10].

3 CITATION COUNTS, PATENT HISTORY, AND

TRADE PRESS ARTICLES

www.sap.com/solutions/sapbusinessobjects/large/busines
s-intelligence/dashboard-visualization/advancedvisualization/index.epx
10 en.wikipedia.org/wiki/Hyperbolic_tree
11 ucjeps.berkeley.edu/map2.html
12 hypergraph.sourceforge.net/examples-general.html

Tracking academic papers would seem to be relatively

easy since these data are extremely important to many
people and there is a long history of analyzing these data
for research projects, tenure decisions, scientometric analyses, and history of science studies. Academic authors
have an obligation to cite relevant previous work which

TABLE 2
SEARCH RESULTS FOR EACH INNOVATION
Google

Google
Scholar

ACM
Digital
Library

HCI
BIB

Web of Science IEEE

Xplore

Cite-seer

USPTO

LexisNexis

Treemap

318,000

7090

132

16

24

57
+
27

602
+
1199

38

112

Cone tree

25,700

1150

19

59
+
396

44

Hyperbolic tree

87,700

1020

20

197

38

45

Frequency counts from various services as of March 20, 2010. The term treemap has a strong showing, but the Google numbers especially are inflated since some
references are to a phylogenetic term, a Java program, and national park maps. The ideas and papers related to cone trees and hyperbolic trees are often cited, but the
term is not always used, giving it an unfairly lower count. The Web of Science results were checked to verify relevance. IEEE Xplore and Citeseer results are for
(treemap) + (tree map) or for (conetree) + (cone tree).

makes tracking influence possible, especially when facilitated by some search services. Similarly, there is strong
interest in patents and patent authors are required to
point to prior art that influenced their innovations. The
U.S. Patent and Trademark Office provides free public
access to the patents, but extracting links to previous patents or academic papers requires diligent effort, since
these links are not well supported. Trade press articles are
available through commercial search services such as Lexis/Nexis and free services such as Googles or Bings
search engine. However, trade press articles rarely link to
or mention academic papers, patents, or related trade
press articles. Instead they name corporations or products, mention key terms or concepts, and quote from key
individuals, but extracting this information is difficult.
An initial study of academic citation patterns for the
six primary papers shows the strongest and most continuing interest in the original cone tree papers (Table 1). An
initial study of use of the terms treemaps, cone trees, and
hyperbolic trees shows the strongest pattern of mentions
for treemaps possibly indicating broader interest beyond
the research community (Table 2). These searches give an
overall impression, but detailed analysis is necessary as
they may include extraneous citations and mentions
while omitting relevant ones. The terms are fairly specific,
but treemaps are also used in phylogenetic research and
as a Java program, while cone trees generate some results
from botanical studies.
To find more detailed, relevant results as well as analyze the trajectory of citations over time we decided to
pursue a more conservative approach. We compiled datasets for the three innovations from three sources. For academic papers we obtained from ACM an XML database
dump of the ACM Digital Library13. This provided us
with metadata for the vast majority of the papers published dealing with these visualization innovations and
should provide a representative sample of the visualization literature. Moreover, it also includes cleaned citations
between papers with approximately 30% of all references
within our post-1990 dataset linked to papers within the

ACM Digital Library. We obtained patent records

through automated scraping of the U.S. Patent and
Trademark Office patent database14, which includes abstracts, full texts, linked citations to other patents in the
database (100% linked), and other references to academic
papers (unlinked). 15. The USPTO database is less usable
than Google Patents for reading individual papers but it
was easier for us to extract references to other patents and
papers from it. Finally, we obtained trade journal and
newspaper articles from the Lexis/Nexis Academic Database using an automated scraping tool. It provides a wide
selection of news, political, legal, business, and reference
information including trade journals, newspaper articles,
and press releases.
Within the ACM Digital Library database dump we
searched for papers where the title, abstract, or keywords
matched one of the three innovations. We called each of
the matching papers flagged papers. This search methodology limited the results to papers that we believed
made contributions to one of the three tree browsing
strategies. The search on title, abstract, and keywords was
performed using these regular expressions:
\b(tree[- ]?map)[s]?\b

\b(cone[- ]?tree)[s]?\b
\b(hyperbolic[- ]?tree)[s]?\b
Each regular expression is composed of a starting
word boundary \b, the first part of the name, e.g.
tree, an optional hyphen or space [- ]? , the
second part of the name, e.g. map, an optional plural
form [s]?, and finally another word boundary. Using
treemap as an example, this regular expression will match
all 6 of the following variations: treemap, treemaps,
tree-map, tree-maps, tree map, tree maps.
This accounted for all but one of the innovation name
variations we discovered within the dump. The last, a
lone instance of hypertree used for the visualization
among a huge number of parallel processing usages, was
added to the flagged papers manually. There were, how14

13

portal.acm.org (subscription required)

15

patft.uspto.gov
www.lexisnexis.com/us/lnacademic (subscription req.)

SHNEIDERMAN ET AL.: INNOVATION TRA

AJECTORIES FOR
R INFORMATION VISUALIZATIONS:
V
ARING TREEMAP
PS, CONETREES, AND
A
HYPERBOLIC
C TREES
COMPA

ever, a handful of critically impo

ortant papers that did not
match
h these restricctions. These included [18][19][28], the
pivottal work on hyperbolic
h
treees and cone trees. These
threee omissions weere quickly discovered due to our prior
wledge of the field, though
h we present a technique
know
later to help identify omitted potentially relev
vant publicat
to the flagged
f
paperrs manually
tions. We added them
onal omitted
after their discovery and checkeed for additio
but reelevant paperss in the dataseet.
Ad
dditionally, we
w added any publications that cited a
flagg
ged paper, ass well as an
ny publication
ns cited by
flagg
ged papers as well. We call these papers alters to the
flagg
ged papers, or just alters. Wee added to ou
ur citation or
edge dataset all cittations (going either directio
on) between
flagg
ged papers, flaagged papers and alters, and
a
between
just aalters. The beenefits of including the altters and the
edgess between them are threefolld. First, it giv
ves us a way
to count the numb
ber of citations each paper receives, inc
into time periodss. Second, it
cluding binning citations
ws us to find commonly
c
citeed papers with
hin an innoallow
vation, providing leads to poteentially relevaant publicam
our searrch terms. Finaally, it gives
tions that didnt match
us a rough idea off the publicatiion and citatio
on trends in
A
any papers to the
the fiield to compaare against. Adding
datasset manually necessitates in
ncluding all citations
c
between
n it and all flaagged papers and
a alters (in either direction),, as well as adding the req
quisite new allters and all
citatio
ons between them
t
and all flagged
f
papers and alters.
This is required to
o ensure prop
per citation co
ounts and to
missed paperss that would have
h
been hig
ghlighted by
find m
the new addition.
arly, though instead of
Paatents were added simila
searching a databaase dump forr the matchin
ng keywords
d
the aautomated scraping tool wee developed downloaded
only the flagged patents
p
and their alters. For matching it
gular expressiions as we ussed for acaused the same reg
demiic papers, searrching across the abstract and
a
full text
of thee patents. Thiis wonderfully
y clean dataseet of patentto-paatent citations was easy to automatically
y extract as
well, including alll citations in either directiion between
flagg
ged patents, flaagged patentss and alters, and
a
between
just aalters.
Th
he links betweeen patents an
nd papers were more difficultt to extract ass only the pla
ain text references section
was aavailable. We wrote a simplle title and yeear matching
tool tto compare th
he references to the cleaneed academic
paperrs dataset whiich was able to
o extract the vast
v
majority
of thee citations fro
om patents to papers. The reverse
r
citations from papers to patents weere more difficcult as there
is a h
huge range of formats for citing patents in
n the papers
in ou
ur dataset16, alo
ong with OCR
R failures in th
he ACM referencces. However,, none of the flagged papers cited any
relevant patents, so
s these edgees are excludeed from the
ysis.
analy
scraped by
Leexis/Nexis articles were automatically
a
16 H
Here is an exam
mple lightly teested regular expression
e
to
match
h most patent citations within the ACM DL
D dump (all
one liine):
((U
United State
es|US|U\.S\.|U\. S\.|U
U\.S) )
?pa
atent(</i>)?
?,? ?(US|No
o[\.]?|Numbe
er)?,? ?
(US
S|#)?[\d, -\
\.]{4,9}\d\b

Fig. 1: Treema
ap visualization for finding highlyy cited hyperbolic tree
papers (green) and patents (grray) by drilling do
own and filtering on an
on attributes ava
ailable.
edge list that has source and target publicatio
o each cell repre
esents
Each cell is a target publication, and the size of
the number off source publicattions that cite it. The most highlyy cited
(largest) float to
t the top left to be ana-lyzed. Here
H
the top three
e publications were
ent properly flag
gged as HT pap
pers and patents. We
created a hierrarchy based on
n the matching innovations or fla
ags of
the source an
nd target pub-lica
ations: SourceFlags -> TargetFla
ags ->
TargetNode. Then
T
we drill dow
wn to the portion of the treemap with a
particular Sou
urceFlag like HT
T (for hyperbolicc tree) and an empty
Target-Flag. Those
T
publication
ns are cited by HT
H papers but are
a not
flagged for anyy innovation.

querying keeywords for different

d
tree structures and
d exporting the results in biblliographic form
mat. It was paarsed
m
it suitablle for visual analya
by automatiic scripts to make
sis. We also manually reaad each articlee and collected the
data about the major orrganizations and
a
productss discussed. Therre were no eassily identifiablle citations to these
articles as even
e
the URL
Ls werent available. Even with
proper URL
Ls, web citatio
on networks are much hard
der to
accurately extract and hav
ve much less metadata attaached
dnt collect an
ny citation datta for
to them. As such, we did
them.
vided
Our conservative apprroach in these searches prov
ow estimate fo
or the total impact of these innous with a lo
vations. Morre liberal stud
dies have largeer numbers, bu
ut for
the purposees of this papeer we wanted
d to identify direct
d
contributorss and the impaact of their co
ontributions. MoreM
over we can
n do a much more
m
careful and
a
accurate analya
sis unclutterred by the chaaff. As an exam
mple, comparred to
the straightfforward basicc search prov
vided by the ACM
A
Digital Libraary, which inccludes referen
nces and review
ws in
the search, our techniquee matched 1332 treemap paapers
instead of 3557, many of which were nott appropriate. Similarly, we fou
und 19 cone tree and 20 hyp
perbolic tree paapers
compared to
o 75 and 100 found onlinee. We were ab
ble to
verify each rejected matcch from the online
o
resultss maugh many of them
t
ended up
u being addeed as
nually, thou
alters due to
o the referencees that the bassic search mattched
against. Thee majority of the alters werent
w
particu
ularly
relevant to the innovation
n, only citing th
he work in a single
s
line or parag
graph. This was
w even moree the case with the
patents, whiich have a ten
ndency to citee even tangen
ntially
related patents and paperrs in an effortt to be compreehensive.
t ensure thatt our search teerms werent omito
In order to
ting any releevant publicaations we anallyzed the resu
ulting
network of publications and citations between theem to

Fig. 2: Tire tracks diagrams showing academic research, industrial

research, and the emergence of $1B commercial markets.

find any publications that were highly cited by publications with a particular flag but did not have that flag
themselves. A treemap is a straightforward visualization
for finding omitted publications by using a hierarchy
based on the citations themselves. We created a hierarchy
with the innovations or flags of the source paper at the
first level, the flags of the target paper at the second, and
finally the target papers themselves at the third and last
level. From that we built the treemap visualization of papers and patents seen in Fig. 1 in Spotfire. It represents
the potentially omitted papers as large rectangles floated
to the top left of the screen, sized by the number of citations matching the source and target flags specified (in
this case, a source matching HT for hyperbolic trees and a
target that doesnt match any of our three search terms
(empty). Here the papers are colored light green and the
patents gray, and the publications with low citation numbers matching the hierarchy have been cut off at the bottom.
Here the top three publications werent properly
flagged as HT papers and patents because they didnt
mention hyperbolic tree anywhere within their title,
abstract, or keywords (for papers) or their full text (for
patents). As it turns out, these three are critical hyperbolic
tree publications that would have substantially impacted
the validity of our results, though we had noted their absence before due to our prior knowledge. We easily found
a critical omitted cone tree paper and patent through the
same technique.
For more details on the complexities of our data cleaning and analysis, see the follow-up paper.

4 VISUALIZING INNOVATION TRAJECTORIES

While there are many tools for showing timelines of numerical data, categorical event data, and historical time
lines, there is relatively little work on showing the multiple complex longitudinal relationships among different
types of data such as academic papers, patents, and trade
press articles. Even more complex is showing the historical pattern of linkages among commercial suppliers,
products, press releases, and adopting organizations.
Visualizations for historical academic citation patterns
is usually traced to Garfields arguments for historiographs to show the impact of a single paper [14][15]. A
more general tool, DIVA shows multiple horizontal time
lines for each topic with nodes showing papers and links
to show citations within and across the time lines [22][23].
The impact of key authors at four human-computer interaction conferences is thoroughly examined and visualized with several forms of networks but less emphasis on
temporal relationships [16]. Citation patterns across sub-

topics are shown in CiteSpace, which recently added the

capacity to locate nodes on timelines [9]. The progress of
scientific arguments is visualized to show opposing opinions and supportive research studies [36]. Attribute
based network visualization which shows temporal patterns for academic papers as well as links across topics
provides an inspiration for our current research [3].
The business literature discusses the process of innovation adoption, especially as measured by trade press articles [39][40] and income from sales [38]. Wang and
Swanson [38] show a 5-year sales histogram, plus growth
rate, and textual annotations of significant events. The
famed Gartners Hype Cycle is a chart with five stages in
the evolving expectations for and visibility of an innovation: Technology Trigger, Peak of Inflated Expectations,
Trough of Disillusionment, Slope of Enlightenment, and
Plateau of Productivity [13]. However, critics complain
that this widely discussed chart is not really a cycle, does
not reflect reality, and fails to provide a basis for action.
A compact visualization that shows relationships
among academic research, industrial research, and the
emergence of billion dollar industries was described as
tire tracks diagrams because of their visual similarity to
the tracks a wet tire might make on pavement [11][20].
These diagrams are appealingly simple and suggest the
strong impact of academic research on commercial success, but they are based on the knowledge of and discussions among selected researchers (Fig. 2). Attempts to
replicate them from public data have not proven successful. Historical time lines of many kinds are a popular method, but there are few strategies that can take data from
multiple and diverse public sources and automatically
create complex timelines that show innovation trajectories.

4.1 Publications
Perhaps the most straightforward way of visualizing the
adoption or impact of each innovation is by using line
charts to show the number of publications for each. Fig 3.,
prepared with Spotfire, shows separate line charts for the
trade press articles (top), academic papers (middle), and
patents (bottom). Each line shows the trajectory for one
innovation, with the publication year on the x-axis and
the number of publications that year on the y-axis. Green
represents treemaps (TM), blue hyperbolic trees (HT), and
red cone trees (CT). A few key features are immediately
apparent. There is a large spike of 17 trade press articles
published for hyperbolic trees in 2000. Drilling down into
the data it became apparent that Inxight was heavily
promoting their hyperbolic tree products with press releases, which accounted for almost all the publications.
Its possible their promotion helped swell hyperbolic trees
in academic papers slightly in the subsequent years.
For both trade press articles and academic papers
treemaps became steadily more mainstream after 2002
and were promoted heavily by the Hive Group, Panopticon, SequoiaView, and others. This increase was matched
by two years of somewhat increased treemap patents,
predominantly assigned to the Hive Group, IBM, and
Microsoft; though the patents tapered off while trade ar-

SHNEIDERMAN ET AL.: INNOVATION TRA

AJECTORIES FOR
R INFORMATION VISUALIZATIONS:
V
ARING TREEMAP
PS, CONETREES, AND
A
HYPERBOLIC
C TREES
COMPA

ticles and academiic papers con

ntinued to incrrease. These
nt figures seem
m to fall off sharply
s
durin
ng 2006-2008
paten
(shad
ded gray). Thiis change wass traced to the time it took
the U
USPTO to pro
ocess patents increased
i
draamatically to
an av
verage of 32 months
m
during
g 2005-2008 (the
(
publica1 . Our datation y
year we use fo
or patents is th
he filing date)17
set off matching paatents and their alters conttains only 1
paten
nt from 2009, compared with 200, 152, 66, and 19 in
2005, 2006, 2007, and
a
2008, resp
pectively. Wee omitted all
o incompleten
ness from all
2009 data from thee charts due to
data sources.
our d
Trreemaps were a little late to
t the patent party, with
cone trees and hyp
perbolic trees already heav
vily patented
from 1997 to 2003. Cone trees weere patented by
b 12 unique
assign
nees led by Xeerox with 18, 41% of the tottal, followed
by Su
un with 7 an
nd Microsoft with 5. Hypeerbolic trees
were less concentrrated by assign
nee with 20 unique
u
assignees, though again
n Xerox led wiith almost a qu
uarter of the
total (9). The otherr top contendeers were IBM
M (4), Rosetta
w
concenInphaarmatics (3), and HP (3). Treemaps were
trated
d similarly with
w
19 uniquee assignees. IB
BM was the
leadeer in treemapss with 12 (31% of the total), with runner
ups M
Microsoft and the Hive Grou
up tied with 4.

4.2 C
Citations
Insteaad of simply looking at the number of publications
p
each year for an inn
novation we can
c analyze th
he number of
ons the related publicationss receive overr time. Fig. 4
citatio
proviides an exam
mple, where th
he y-axis reprresents how
many
y times match
hing publicatiions have beeen cited per
year instead of showing the nu
umber of publlications per
year. Unfortunattely there arre no easily measurable
citatio
on data for traade journal arrticles so they are omitted
in furrther analysess. Separate liine charts sho
ow academic
paperrs (top) and patents
p
(bottom
m), with the publication
a line color green for treeemaps (TM),
year on the x-axis and
nd red for con
ne trees (CT).
blue ffor hyperbolicc trees (HT), an
The y
y-axis is the co
ount of all cita
ations from bo
oth academic
paperrs and patentts published that
t
year (srcc) where the
cited publication matches
m
the in
nnovation. Wee include patent-tto-paper citatiions in the cittation count fo
or academic
paperrs, though no
o paper-to-pa
atent citations existed for
term--matching pattents in our da
ataset.
Th
his visualizatio
on paints a so
omewhat diffeerent picture
for ccone trees an
nd hyperbolic trees, wheree instead of
shortter spurts of academic
a
papeers and patentts published
theree is a more prolonged and sizable measu
ured impact.
Even
n though theree were only a few cone treee academic
paperrs published each year, theey were cited 44 times in
1998 and 55 in 2002.
2
Citation
ns for cone tree
t
patents
peakeed in 2000 (194) and for hyp
perbolic trees later in 2005
(144),, also a local maxima for cone
c
trees (153). Treemap
paten
nts were nev
ver comparatively highly
y cited but
peakeed 2003-2004 with 45. Alll have been following a
stead
dy descent in the period of 2006-2008 (sh
haded gray),
thoug
gh this match
hes closely thee reduction in
n patents in
our d
dataset during
g that time peeriod, possibly
y due to the
backllog at the USP
PTO.
17
www.ipwa
atchdog.com/2
2009/04/22/usp
pto-backlogpaten
nt-pendency-o
out-of-control/id
d=2848

Fig. 3: These line charts show the number of

o trade press articles
a
mic papers (midd
dle), and patents (bottom) pub
blished
(top), academ
each year mattching each innovvation. Each line
e represents a diffferent
innovation: tre
eemaps (TM/gre
een), hyperbolic trees (HT/blue)), and
cone trees (C
CT/red). Note: th
he sharp fall in patent figures in the
faded area ma
ay be to average
e 32-month USPTO processing time in
2005-2008.

Fig. 4: Thesee line charts shoow the number of citations acaademic

papers (top) and patents (boottom) matchingg each innovatioon rey
Each line reepresents a diffeerent innovationn: treeceived each year.
maps (TM/grreen), hyperboliic trees (HT/bllue), and cone trees
(CT/red). Notte: the sharp falll in patent figuures in the fadedd area
may be to avverage 32-monthh USPTO processing time in 20052008.

Fig. 5: The total numbeer of citations eaach academic paaper and pah
providing another measuree of its influtent receives is plotted here,
ence. The number of ciitations, or in-deegree, a publicattion receives
p
year is plotted on
is plotteed on the verticaal axis and the publication
the horrizontal. The rows
r
divide th
he innovations: cone trees
(CT/topp), hyperbolic treees (HT/middle), and treemaps (TM/bottom)
(
and thee columns publiccation type: academic papers (lleft) and patents rigght).

Ano
other approach
h to measure adoption or im
mpact is to
dig do
own to the level of individu
ual papers and
d patents to
see th
heir influence.. The scatterp
plots in Fig. 5 show a
trianglle for each pu
ublication with
h the y-axis reepresenting
the to
otal number of citations it received from
f
both
academ
mic papers and patentss and the horizontal
publicaation year. Th
he left column
n shows acadeemic papers
and th
he right patentss, with a row for
f each innovation: cone
trees ((CT/top), hyp
perbolic trees (HT/middle),, and treemaps (TM/bottom). Here the keey publication
ns for each
innovaation are appaarent, with lab
bels where therre is space.
The 19991 ACM CH
HI paper on cone trees [29] has been
cited aan amazing 2442 times, with
h the 1993 CA
ACM paper
[28] cited a respectaable 132. Thee first patent in
i 1993 by
Roberttson et al., Display of hierarchiccal threedimen
nsional structu
ures with rotatting substructtures, was
cited 1130 times and the two most cited followu
ups came in
1997: P
Pirolli et al., System

for categorizing docu

uments in a
linked collection of documents (1
138) and Pirolli et al.,
System
m for predictingg documents reelevant to focuss documents
by spreeading activatioon through nettwork representations of a
linked ccollection of documents (103). Those only only touch
on con
ne trees as an application
a
of their patent, th
hough.
For hyperbolic trrees, the 2-pag
ge 1994 UIST description
d
[18] gaarnered 47 citaations, though
h dwarfed by the 156 for
the 19995 ACM CHI paper [19]. The
T enhanced version by
the Un
niversity of Waterloo
W
[8] recceived only 30 citations.
There is an interesting mix of hig
ghly-cited pateents during
1994-2000. There weere not any highly
h
cited trreemap pa-

Fig. 6: The citaation trajectory of

o the two top ciited academic paapers
and the two pattents with the most
m consistent or
o interesting citaation
patterns is show
wn here. Each colored
c
line reprresents a publicaation
and indicates how
h
many citatioons it received each
e
year. This provides another view
v
of their lonng-term impact. The rows dividee the
innovations: cone trees (CT/topp), hyperbolic treees (HT/middle),, and
treemaps (TM/bbottom) and the columns publicaation type: acadeemic
papers (left) annd patents right)). Note: the sharrp fall in patentt figures in the faaded area may be to average 32-month USP
PTO
processing timee in 2005-2008.

tents, though
h there was an
a early 1993 patent by Baaker
with 27 and the
t original paapers did quitte well. The 1991
1
IEEE Vis pap
per [16] was cited 158 tim
mes and the 1992
1
ACM TOG paaper [28] got 143. A notable departure frrom
the other inno
ovations is thee highly cited
d 2001 PhotoM
Mesa
paper and tw
wo new design
n papers in 19999 and 2002 that
may have help
ped usher in a renewed inteerest in treemaaps.
For an even
n finer-grained look at the citation
c
history of
individual publications
p
w can plot the numberr of
we
citations each
h receives perr year as a lin
ne chart (Fig.. 6).
There each lin
ne represents an individuall publication, and
its height at each year ind
dicates how many
m
citation
ns it
n, columns arre by publicattion
received thatt year. Again
type (academ
mic papers on th
he left and pattents on the rig
ght)
with innovatiions separated
d by row: con
ne trees (CT/top),
hyperbolic
trees
(HT
T/middle),
maps
and
treem
(TM/bottom). The cone trree papers an
nd the hyperb
bolic
he top with quick
q
uptake and
tree papers stand out at th
map
relatively horrizontal trajectories, while the two treem
papers start from humbleer beginnings but consisten
ntly
g between hyp
perbolic trees and
a cone treess for
climb, ending
total citationss. Both the cone
c
tree and
d hyperbolic tree
patents are highly cited over time, while the main
m
nt was consisttently but nott frequently ciited.
treemap paten
The 2003 treeemap patent by
b Zlatanov and
a
Furlong iss an
abberation, highly
h
cited in 2004 but back
b
to previious
norms in 20005. It will be interesting
i
to see its trajecttory

SHNEIDERMAN ET AL.: INNOVATION TRAJECTORIES FOR INFORMATION VISUALIZATIONS:

COMPARING TREEMAPS, CONETREES, AND HYPERBOLIC TREES

over the next decade.

5 DISCUSSION
Our sample of these three information visualizations for
tree structures is a narrow one, so generalizations are difficult, but some topics invite discussion.
Some new ideas catch on rapidly and spread widely,
but most fail, while another group takes decades of
steady refinement and tuning to fit appropriate application domains. The entrenched command line interface
slowly gave way to the graphical user interface, which
has opened the doors for many users. Similarly, the entrenched textual, numeric, and tabular interfaces for data
are giving way slowly to new information visualization
ideas. Change is possible, although new application domains, platforms, tasks, and users may offer fertile possibilities, much as Moore [21] and Ogle [25] suggest. Rao
comments on these issues in his Information Flow blog in
2003, but even his prediction for 2007 has not yet been
realized18.
These three cases suggest that an important promoter
of adoption is having a readily understandable demonstration using a familiar application domain, while carrying out commonly used tasks. Treemaps started with directory browsing, but became a broader success when
other appropriate applications appeared, such as Smartmoneys MarketMap, supply chain management, insurance fraud, stock portfolio analysis, and production management. Cone trees began with directory browsing and
spread to web browsing, but few other applications.
Hyperbolic trees were used for information-rich applications such as organization charts, web browsing, and
document libraries.
Both sides of the long-standing controversial issue of 2dimensional vs. 3-dimensional information visualization
can make points based on tree browsing. While several 3dimensional treemaps were implemented, they quickly
faded and empirical evidence showed few benefits over
2-dimensional versions. The 3-dimensional cone tree
generated strong interest and may have inspired many
developers, but the lack of commercial diffusion raises
questions about the complexity of using 3-dimensional
representations for these tasks.
Enthusiasts for user-controlled animated visualizations
included the designers of cone trees and hyperbolic trees,
which were sufficiently appealing to generate strong interest but the modest commercial diffusion raises questions of efficacy. Does the frequent change of position of
nodes undermine recall and discovery? By contrast the
treemaps had limited animation, mostly related to zooming in on a subtree, but until recently, this was usually
done with a simple jump zoom rather than a smooth
animation. Researchers on treemaps focused on spatial
stability, especially as node sizes changed, to prevent the
disturbing movement and reshaping of rectangles. Later
refinements to treemaps addressed these issues but have
18

www.ramanarao.com/informationflow/archive/200302.html

yet to appear in commercial versions.

While Shneiderman has written and spoken in favor of
intellectual property protection, these cases add support
to those who argue that patents inhibit adoption and diffusion. The unpatented treemaps led to many implementations with refinements and a lively competition among
commercial and open source developers. By contrast, the
patented cone trees and hyperbolic trees limited the
number of derivative implementations. Ironically, the
number of treemap refinement patents approximately
matches the number of patents for cone trees and hyperbolic trees. Card points out the further irony that the
academically-oriented University of Maryland produced
a more broadly adopted and commercialized technology,
while the industrially-oriented Xerox PARC produced
strong and more broadly cited papers (personal communication 3/30/2010).

CONCLUSION

Understanding the innovation trajectory for novel research ideas from invention to successful innovation is a
difficult task. Even when there are evaluations of efficacy
and precise records of early adoption it seems difficult to
predict durability. Academic citation counts are not always tied to patent counts or trade press articles, and certainly many factors govern commercial success. A few of
the determinants of success are the entrenched alternatives, resistance to change, interactions with existing
technologies, intellectual property rights, perceived usefulness and ease-of-use, availability of free versions, influence of entrepreneurial individuals, and responsiveness of trade press editors. This complex mix of personalities, ideas, institutions, and economic constraints makes it
difficult to predict outcomes even 2-3 years in the future.
On the more positive side, readers might take away an
appreciation of how many components there are to commercial success and the sobering realization of how much
effort is required in many cases. Early evaluations of efficacy may help lead to refinements, as well as a clearer
understanding of which tasks and application domains
are a good match for an information visualization invention. Presentations at conferences, through videos, and at
academic seminars can provide further feedback and a
chance to win over visionaries and early adopters. The
shift to a commercial effort is still more demanding,
probably requiring enthusiastic champions (as Ramana
Rao certainly was for Inxights hyperbolic trees), a devoted team of developers, and a determined sales force.
Small or large companies can help advance an invention
to a broadly used innovation.
There are also many dimensions to success. Academic
success is typically measured by citation counts which are
presumed to be an indicator of influence on other researchers. Commercial success could be measured in dollars, number of customers, or numbers of users. For inventors, there is great satisfaction and pride in seeing
their ideas influencing others and being put to use for
important applications. The information visualization
community faces substantial challenges in bringing their

11

ideas to broad adoption, but each success creates a greater

visual literacy, which sharpens users critical thinking
while making them still more sympathetic to the next
innovation.

[17]

ACKNOWLEDGMENT

[18]

The authors wish to thank Jim Bartoo, Stu Card, Jock

Mackinlay, Ramana Rao, and George Robertson for their
comments on the history, Catherine Plaisant for comments on drafts, and the ACM for supplying us with a
dump of their digital library. This work was partially
supported by National Science Foundation grant SBE0915645.

REFERENCES
[1]
[2]

[3]

[4]
[5]
[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]
[15]

[16]

Allen, T. J., Managing the Flow of Information Technology, MIT

Press, Cambridge, MA (1977).
Allen, M. M., The hype over hyperbolic browsers, ONLINE 26, 3
(May/June
1992),
Available
at
http://www.onlinemag.net/may02/allen.htm.
Aris, A., Shneiderman, B., Qazvinian, V., and Radev, D., Visual
overviews for discovering key papers and influences across research
fronts, Journal of the American Society for Information Systems and
Technology 60, 11 (November 2009), 2219-2228.
Arthur, B., The Nature of Technology: What It Is and How It Evolves,
Free Press, New York, (2009).
Basalla, G., The Evolution of Technology, Cambridge University
Press, New York (1988).
Bederson, B. B., Shneiderman, B., and Wattenberg, M., Ordered and
quantum treemaps: Making effective use of 2D space to display hierarchies, ACM Transactions on Graphics (TOG) 21, 4 (October 2002),
833-854.
Bruls, M., Huizing, K., & van Wijk, J. J., Squarified treemaps. Proc.
Joint Eurographics and IEEE TCVG Symposium on Visualization
(TCVG 2000) IEEE Press, Piscataway, NJ (2000), 33-42.
Carriere, J. and Kazman, R., Research report: Interacting with huge
hierarchies: beyond cone trees, IEEE Symposium on Information Visualization (InfoVis '95), IEEE Press (1995), 74-81.
Chen, C., CiteSpace II: Detecting and visualizing emerging trends and
transient patterns in scientific literature. Journal of the American Society for Information Science and Technology, 57, 3 (2006), 359-377.
***ADD*** Chen, C., Ibekwe-SanJuan, F., & Hou, J., The structure
and dynamics of co-citation clusters: A multiple-perspective cocitation analysis. Journal of the American Society for Information
Science and Technology, 61, 7 (2010), 1386-1409.
Cockburn, A., Karlson, A., and Bederson, B. B., A review of overview+detail, zooming, and focus+context interfaces, ACM Computing
Surveys 41, 1 (December 2008), Article No. 2.
Computer Science and Telecommunications Board, Innovation in
Information Technology. National Academies Press, Washington, DC
(2003).
http://www7.nationalacademies.org/cstb/pub_itinnovation.html
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and
user acceptance of information technology. MIS Quarterly, 13, 3
(1989), 319-340.
Fenn, J. and Raskino, M., Mastering the Hype Cycle: How to Choose
the Right Innovation at the Right Time, Harvard Business Press, Boston, MA (2008).
Garfield, E., Historiographic Mapping of Knowledge Domains Literature, Journal of Information Science 30, 2 (2004), 119-145.
Garfield, E., Pudovkin, A. I., and Istomin, V. I., Why do we need
algorithmic historiography? Journal of the American Society for Information Science and Technology (JASIST) 54, 5 (2003), 400412.
Henry, N., Goodell, H., Elmqvist, N., and Fekete, J.-D. (2007). 20
Years of Four HCI Conferences: A Visual Exploration, Intl Journal

[19]

[20]
[21]
[22]
[23]

[24]

[25]
[26]

[27]

[28]

[29]

[30]
[31]

[32]

[33]

[34]

[35]

[36]

of Human-Computer Interaction, 23, 3(2007), 239-285, Available at:

http://dx.doi.org/10.1080/10447310701702402
Johnson, B. and Shneiderman, B., Tree-maps: A space filling approach to the visualization of hierarchical information structures,
Proc. IEEE Visualization 91, IEEE Press, Piscataway, NJ (October
1991), 284-291.
Lamping, J. and Rao, R., Laying out and visualizing large trees using
a hyperbolic space, Proc. 7th Annual ACM symposium on User Interface Software and Technology, ACM Press, New York (1994), 13-14.
Lamping, J., Rao, R., and Pirolli, P., A focus+context technique based
on hyperbolic geometry for visualizing large hierarchies, Proc. SIGCHI Conference on Human Factors in Computing Systems, ACM
Press, New York (1995), 401-408.
Lazowska, E., Envisioning the future of computing research, Communications of the ACM 51, 8 (August 2008), 28-30.
Moore, G. A., Crossing the Chasm, HarperCollins, New York, 1991.
Morris, S. and Martens, B., Mapping Research Specialties, Annual
Review of Information Science 42 (2007), Chapter 6.
Morris, S. A., Yen, G., Zheng, W., and Asnake, B. (2003). Time Line
Visualization of Research Fronts. Journal of the American Society for
Information Science and Technology (JASIST) 54, 5 (2003), 413
422.
Mullet, K., Fry, C., and Schiano, D., On your marks, get set, browse!
(The Great CHI97 Browse Off), ACM CHI97 Conference Extended
Abstracts, ACM Press, New York (1997), 113-114. Available at:
http://www.sigchi.org/chi97/proceedings/panel/kem.htm
Ogle, R., Smart World: Breakthrough creativity and the New Science
of Ideas, Harvard Business School Press, Boston, MA (2007).
Pirolli, P., Card, S., and Van Der Wege, M., The effects of information scent on visual search in the hyperbolic tree browser. ACM
Trans. Computer-Human Interaction 10, 1 (2003), 2053.
Plaisant, C., Grosjean, J., and Bederson, B., SpaceTree: Supporting
exploration in large node link tree, design evolution and empirical
evaluation. Proc. IEEE Symposium on Information Visualization (InfoVis), IEEE Press, Piscataway, NJ (2002), 5764.
Robertson, G. G., Card, S. K., and Mackinlay, J. D., Information
visualization using 3D interactive animation, Communications of the
ACM, 36, 4 (1993), 51-71.
Robertson, G. G., Mackinlay, J. D., and Card, S. K., Cone trees: Animated 3D visualizations of hierarchical information, Proc. ACM
SIGCHI Conference on Human Factors in Computing Systems, ACM
Press, New York, (April 1991), 189-194.
Rogers, E. M., Diffusion of innovations (5th ed.). New York: Free
Press, New York (2003).
Shneiderman, B., Tree visualization with tree-maps: A 2-dimensional
space filling approach, ACM Transactions on Graphics 11, 1 (January
1992), 92-99.
Shneiderman, B. and Plaisant, C., Strategies for evaluating information visualization tools: Multi-dimensional In-depth Long-term Case
Studies, In Proc. Beyond time and errors: novel evaluation methods
for Information Visualization, Workshop of the Advanced Visual Interfaces Conference, Available in ACM Digital Library (2006), 1-7.
Shneiderman, B. and Plaisant, C., Designing the User Interface: Strategies for Effective Human-Computer Interaction: Fifth Edition, Addison-Wesley Publ. Co., Reading, MA (2010).
Stasko J, Catrambone R, Guzdial M, and McDonald, K., An evaluation of space-filling information visualizations for depicting hierarchical structures, Intl J. of Human-Computer Studies 53, 5 (Nov.
2000), 663-694.
Turo, D. and Johnson, B., Improving the visualization of hierarchies
with treemaps: design issues and experimentation, Proc. IEEE Conference on Visualization 1992, IEEE Press, Piscataway, NJ (1992), 124131.
Uren, V., Buckingham Shum, S., Bachler, M., and Li, G. (2006).
Sensemaking tools for understanding research literatures: Design, implementation and user evaluation, Intl Journal of Human-Computer
Studies 64, 5 (2006), 420-445.

SHNEIDERMAN ET AL.: INNOVATION TRAJECTORIES FOR INFORMATION VISUALIZATIONS:

COMPARING TREEMAPS, CONETREES, AND HYPERBOLIC TREES

[37] Venkatesh, V., Morris, M. G., Davis, G. B. and Davis, F. D., User
acceptance of information technology: Toward a unified view, MIS
Quarterly 27 (2003), 425-478.
[38] Wang, P. and Swanson, E. B., Launching professional services automation: Institutional entrepreneurship for information technology innovations, Information and Organization, 17, 2 (2007), 59-88.
[39] Wang, P. and Ramiller, N.C., Community learning in information
technology innovations, MIS Quarterly, 33, 4 (2009), 709-734.
[40] Wang, P., Popular concepts beyond organizations: Exploring new
dimensions of information technology innovations, Journal of the Association for Information Systems, 10, 1 (2009), 1-30.
[41] Wattenberg, M., Visualizing the Stock Market. Proc. Extended Abstracts of Human Factors in Computing Systems (CHI 99) ACM
Press, New York (1999), 188-189.
[42] Xiang, Y., Chau, M., Atabakhish, H., and Chen, H., Visualizing criminal relationships: Comparison of a hyperbolic tree and a hierarchical
list, Decision Support Systems 41, 1 (Nov. 2005), 69-83.

13